Why Stripe may pay $10B for a router everyone is building

Stripe may be buying the toll booth on inference.

It is reportedly in talks to acquire OpenRouter for around $10 billion, which sounds like a lot for a router at the exact moment everyone is building one.

Cursor has one. Runway has one. Ramp has one. Amazon Bedrock already routes between models. Every serious AI product will probably do the same internally.

The router is not scarce. The position in the flow is.

The router paradox

Standalone routers have not had an easy time.

TensorZero raised funding, attracted thousands of GitHub stars and is now no longer maintained. Helicone built a gateway spanning more than 100 models, then joined Mintlify and moved into maintenance mode.

Routing is useful, but it gets absorbed. Into a product, a cloud, an API gateway or an infrastructure provider.

OpenRouter may be different because developers do more than route requests through it. They discover models there, test them, keep credits and send production traffic through the same integration.

OpenRouter owns the AI usage flow, not only the routing decision. That is the flow everyone wants to get their hands on.

I run a billing company, so I spend a lot of time looking at usage events. Historically, usage arrives after the product action. It gets metered, priced and eventually added to an invoice.

With AI, that can be too late.

Imagine a customer has almost exhausted its credits and sends a request to an expensive model. The product can still downgrade the model, require approval, ask for a top-up or block the request. Once the inference has happened, the cost already exists.

AI usage is not only something to bill. It changes what the product should allow next.

The live path

The companies closest to live AI usage start from different positions.

Amazon Bedrock already routes prompts between models to optimize quality and cost. CoreWeave now exposes inference gateways, deployments and capacity through its API. Nebius has moved from GPU infrastructure into production inference with Token Factory. They see the request and its underlying infrastructure cost.

Snowflake sits somewhere else. It sees AI workloads beside the customer’s data, governance and existing consumption. Its cross-region inference already routes requests according to capacity and geography.

Cursor and Runway see the user’s intent. They can choose a model based on the task and change the product experience immediately. Their limitation is distribution: the router belongs to their own product.

Ramp starts from the budget. Its spend products explain where the money went, while Ramp Router moves it into the live request. That makes Ramp complementary to the platforms managing customer entitlements and monetization.

Kong is closing the loop from API infrastructure. Kong AI Gateway routes requests, identifies consumers and applies token or cost limits. Its acquisition of OpenMeter adds real-time metering, entitlements and billing.

A gateway becomes more useful when it knows who is making the request, what their plan allows and how much usage remains. It becomes more valuable when it can monetize that usage and participate in the money flow.

Everyone is moving toward the same point from a different direction: the moment AI usage can still change what the product allows and what the customer pays.

Business models as code

Stripe already knows the customer, payment method, invoice and credit balance. OpenRouter would give it the event that creates the cost.

The full sequence becomes visible: request, model, cost, entitlement, price, payment.

OpenRouter’s neutrality is part of why customers trust it. They get optionality between models and leverage with providers. Once the router belongs to a company that also owns the billing and payment rails, neutrality becomes harder to separate from strategy.

Integration is another question. Six months after acquiring Metronome, Stripe Billing still has its own usage meters, while Metronome has its own documentation, data model and Stripe App. The products seem to be enjoying a respectful period of getting acquainted.

Stripe and Advent are also reportedly considering a $53 billion offer for PayPal, which comes with roughly 20,000 employees. They seem to have plenty to integrate already.

Stripe helped turn payments into code. AI is turning business models into code.

Each request can execute a different model choice, cost, entitlement and price inside the product. OpenRouter sees the request. Stripe sits in the money flow.

That connection is what could be worth $10 billion.